Reprinted from http://blog.csdn.net/enoch520/article/details/38980129
one What is the block storage

For ordinary users, block storage both familiar and unfamiliar, so to speak, in fact, we usually use PC hard disk, U disk, mobile hard disk can be classified as a storage device, simple, block storage is to provide a block-level storage interface can be persisted data storage devices, including personal PC hard disk, U disk, enterprise-level SAS hard disk, SAN Shared storage, and with the arrival of cloud computing elastic distributed block storage service, etc., and AMAZON EBS is a typical representative of the elastic distributed block storage service, and so far the size scale biggest elastic distributed storage service,

two What is the EBS

EBS full Elastic Block Store, is to provide the AWS block-level storage service, for the AWS EC2 provide independent storage volume, EBS on EC2 appears like a regular hard, but in fact docking is EBS background of distributed storage system, rather than the local hard disk,

AWS EBS provides a very cloudy computing features, such as storage and computation, flexible extensible, a copy of the data, providing a snapshot and quick recovery services, custom and so on, the performance, so to speak, these features are traditional PC local disk is unable to provide, the following will discuss these characteristics to analysis,

3. The feature analysis of the cloud era block storage

1. The separation of computation and storage (elastic)

"Elastic" is the most often mentioned in the cloud computing is a word, the so-called elasticity, and is not fixed, not binding, defined by the software, so the traditional block based on the local disk storage is unable to provide flexible function, because the local disk and the server is bound together by hardware level interface, storage resource and computing resource integration on a motherboard, David and Jonathan, imagine that you buy one in the AWS EC2, also bought a block storage based on the ground mount to EC2, when the EC2 the downtime of machine tools, or hardware damage, EC2 is not available, at the same time block couldn't find the data on the storage device, of course, AWS engineer students can be hard on this server will be removed, change to another server on the data read out, but the operating costs will be very high,

EBS separation by storage and calculation, the realization of the "elastic" characteristic of the block storage, EC2 only provides computing resources, EBS by separately, and then mounted on EC2, EC2 and EBS specific life cycle, independently on EC2, can increase, delete, migration EBS, when the EC2 is unavailable, can in a short time the EBS mount new EC2, continue to use the data completely is not affected, in addition to the fault fast recovery, EBS can also support data migration quickly, the elastic function such as capacity expansion and rapid,

2. The data reliability

For most people, piece of equipment is very high reliability storage device, we use the personal PC, using usb, rarely considering the hard disk, the data on the usb drive will be lost, will not wrong? We write a word document on a PC, only press CTRL + s, we were at ease, in fact, PC hard disk, U disk reliability is really good, an ordinary hard disk, used for several years all have no problem, not to mention the enterprise-level SAS hard disk, SAN storage,

However, for a have thousands, tens of thousands of sets of data center, the reliability of common hard disk? It is actually a probability problem, it has been proved that after the data center to reach a certain size, will be a hardware failure normalized, mainboard failure, hard disk failure, memory failure is almost happen every day, which is responsible for the data in the hard disk, the impact of failure is one of the biggest, and EBS is much x86 server combined into a distributed cluster to provide services, there will be more copies stored data, single server outage or single disk is damaged, will not affect the user data and services, completely transparent to users, so, in terms of probability provided by the EBS block storage service, its data reliability is higher than the traditional local disk,

3. The service availability

For distributed block storage service, maintain service high availability is the biggest challenge, however to block storage, the user is very sensitive to the requirement of availability, can imagine, when your personal PC hard disk suddenly cannot write/read, presumably is going to be very worried, for distributed storage, usability challenges mainly from three aspects, one is the availability of server node, the second is the usability of the network, 3 it is operational complexity,

Availability of server node can copy more, eliminate single point to achieve; And network availability is more complex, as far as I know, EBS master for network is adopted to guarantee the network availability, the main network data services at ordinary times, in the main network anomalies, enabled for web services, and operational complexity, due to the participation of people, so the most uncontrollable, is also the most vulnerable, AWS EBS several failures in the past, most of them are operational error, so it is particularly important to operations automation,

4. Performance

Block storage services directly to users, so the performance is good or bad is extremely sensitive, IO delay 1 ms and IO 10 ms to the user's experience, completely distributed block storage service performance mainly depends on the time delay of two parts, one is the network time delay, it is the performance of the storage medium itself, in the blocks of distributed storage service, due to realize the separation of storage and computing, must introduce the network time delay, in some IO scenarios (such as order, speaking, reading and writing), distributed storage of IO delay than the local disk IO, so in general, the compute nodes and storage nodes deployed across the room, or the effect of network delay will is bigger,

As the cloud computing provider, AWS EBS performance may not be the best, but the EBS is doing better for performance to the user to define, chosen by the user, let the user had a expectations for performance, AmazonEBS provides two types of volume, namely the standard volume and pre-configured IOPS volumes, the performance characteristics of them and the price is different, can be customized according to the requirement of the application and budget required storage performance,

Standard volume may be requires moderate or sudden type of I/O provide storage application, these can provide an average of about 100 IOPS, up to hundreds of IOPS, standard volume is also very suitable for used as a guide, its breaking capacity can provide rapid instance startup time (usually more than 10 seconds),

Pre-configured IOPS volumes to read and write I/O intensive, such as database, provide high performance can be expected workload, create a volume, use the preset IOPS for volume IOPS rate is determined, then AmazonEBS in life cycle with the rate of the volume, AmazonEBS currently support each preconfigured IOPS volume up to 4000 IOPS, you can use multiple banding type roll together, for the application's ability to provide each AmazonEC2 thousands of IOPS,

With the reduction of cost of SSDS AWS is introduced based on EBS SSD media, there will be a promotion on performance, of course the price also will improve,

5. User isolation

Traditional mode of the local disk is single tenant, a disk used by a user exclusive, in the blocks of distributed storage, is multi-tenant sharing a distributed cluster, here the user isolation and flow control is important, a distributed block storage cluster tend to carry hundreds of thousands of users to access, but the total bandwidth of a cluster is limited, if a user for high bandwidth of IO, most of the cluster bandwidth resources, so will inevitably attract other user complaints, so in the blocks of distributed storage, users have to be certain resources isolated and resource constraints, the bandwidth, IOPS critical performance data such as real-time adjustment and control, to guarantee the stable operation of the whole cluster,

Four. The future of distributed block storage

Under the scenario of cloud computing, the use of distributed block storage is the path to realize some of the characteristics of cloud computing, along with the development of the hardware, the architecture of distributed block storage will evolve, such as distributed block storage based on SSD media will be more and more accepted by users; Wan Zhao the development of the network popularization, also can let a step on the performance of distributed block storage, in addition to the AWS, ali cloud cloud for such business, such as the open source community distributed block storage will also continue to develop, such as open source sheepdog, CEPH system, such as open source distributed storage system is very good, some of his design is very ingenious and practical,